maes <- read.table("babies.txt", header=T, sep="")
head(maes)
#maes[maes$gestation==9, 2]<-NA #transformando em NA
#maes[maes$gestation==99, 2]<-NA
#maes[maes$gestation==999, 2]<-NA
#maes[maes$parity==c(9,99,999), 4]<-NA
maes2<-subset(maes, bwt!=999&gestation!=999&parity!=9&age!=99&height!=99&weight!=999&smoke!=9)
#removi as linhas que tinham9, 99, 999, ! e, se fosse ou  |
maes2
head(maes2)
maes.1<-lm(bwt~gestation, data=maes2)
anova(maes.1)
maes.2<-lm(bwt~gestation + parity, data=maes2)
anova(maes.1, maes.2)
plot(maes2, panel=panel.smooth)

maes.3<- lm(bwt~age, data=maes2)
anova(maes.2, maes.3) #no diferem, o menor  melhor.
maes.4 <- lm(bwt~height, data=maes2) 
anova(maes.2, maes.4)
maes.5<- lm(bwt~(gestation + parity+ height)^2, data=maes2)
summary(maes.5)
#para que ele faa os par a par de uma vez vc coloca a varivel resposta e coloca dps do ~ as variveis que vc quer e eleva pelo 2 se for par a par, se for 3 ao cubo.

maes.6<- lm(bwt~gestation+parity+smoke, data=maes2)
summary(maes.6)
anova(maes.2, maes.6)
#Melhor modelo! 
anova(maes.6)
maes.7<- lm(bwt~gestation+smoke, data=maes2)
anova(maes.6, maes.7)

maes.8<- lm(bwt~gestation+height+smoke, data=maes2)
anova(maes.6, maes.8)

maes.9<- lm(bwt~gestation*height+smoke, data=maes2)
anova(maes.8, maes.9)

maes.10<- lm(bwt~gestation+height+gestation:smoke, data=maes2)
anova(maes.6, maes.10)
#maes.10 no tem diferena com maes.6 (que eu considerei melhor modelo).